Back to home

Research Contributions

A deeper dive into the problems I care about (accessibility, security, and AI forensics) and the work I’ve done across papers, benchmarks, and a granted patent.

H-Index: 3•Total Citations: 21+•Google Scholar

Timeline

Federated learning schematic

Rethinking Data Integrity in Federated Learning: Are we ready?

2022
IEEE International WIE ConferenceFederated LearningSecurity•
IEEE Xplore

S Dixit, PN Mahalle, GR Shinde

A security-focused analysis of federated learning attack surfaces, with emphasis on poisoning and tampering risks and practical integrity safeguards.

Contributions

  • Surveyed attack surfaces and real threat scenarios for distributed learning.
  • Analyzed integrity and poisoning risks across aggregation and client updates.
  • Proposed protocol-level mitigations and documented deployability trade-offs.

Abstract

Investigates vulnerabilities in distributed learning—especially poisoning and data tampering—and proposes protocols to improve integrity in federated aggregation.

Hand reading Braille

Assistance Platform for Visually Impaired Person using Image Captioning

2023
Indian Patent OfficePatent GrantedComputer VisionAccessibility

Inventor: Shreyas Dixit

A real-time multimodal narration system that turns visual context into natural language and audio to improve independence and safety for visually impaired users.

Contributions

  • Designed an end-to-end pipeline for capture → captioning → speech in real time.
  • Optimized for latency, clarity, and robustness across varied everyday environments.
  • Filed and secured patent protection (Patent No. 202321004399).

Abstract

The platform converts visual information into descriptive audio via image captioning, enabling users to understand surrounding scenes and objects through hands-free narration.

Ocean waves

Wave-Former: Lag Removing Univariate Long Time Series Forecasting Transformer

2024
Ocean Engineering (Elsevier), Vol 312TransformersTime-Series

Shreyas Dixit, Pradnya Dixit

A Transformer architecture tailored for long-horizon wave forecasting, designed to reduce phase-shift (lag) errors that hurt downstream planning and control.

Contributions

  • Focused on modeling choices that prioritize temporal alignment over only point-wise accuracy.
  • Helped frame experiments for chaotic natural signals where timing errors matter materially.
  • Co-authored the end-to-end narrative from problem definition to evaluation and takeaways.

Abstract

Designs a Transformer architecture to reduce lag (phase shift) in long time series forecasting, improving usability for ocean wave prediction where timing alignment is critical.

Fact Check Explorer screenshot

DeFactify 4: Counter Turing Test (Text & Image)

2024
Workshop ProceedingsWorkshopDatasetBenchmarking

R Roy, G Singh, A Aziz, S Dixit, et al.

Workshop overview and dataset releases for benchmarking human vs. AI-generated text and images with reproducible evaluation protocols.

Contributions

  • Co-authored overview + dataset papers and positioned the benchmark within the AI-forensics landscape.
  • Worked on task definitions and documentation to improve reproducibility and fair comparisons.
  • Supported reporting standards and practical evaluation guidance.

Abstract

Co-authored papers for the DeFactify 4 workshop, establishing benchmarks for human vs. AI-generated text and image detection via overview and dataset releases.

Example photo with a visible watermark

Peccavi: Visual Paraphrase Attack Safe and Distortion Free Image Watermarking

2025
arXiv PreprintGenAIForensicsSecurity•
arXiv

S Dixit, A Aziz, S Bajpai, V Sharma, A Chadha, V Jain, A Das

A watermarking technique for AI-generated images that stays detectable under subtle, semantics-preserving edits while keeping the source visually unchanged.

Contributions

  • Framed the threat model around real attacker behavior: small edits that preserve semantics but break detectors.
  • Contributed to method design and writing around robustness vs. imperceptibility trade-offs.
  • Helped communicate the evaluation story and practical adoption constraints.

Abstract

Proposes a watermarking approach that is robust against "visual paraphrase" attacks—subtle distortions designed to evade traditional detection—while maintaining zero visual distortion in the source media.

Presentation slide asking whether content is real or fake

The Visual Counter Turing Test (VCT²): A Benchmark for Evaluating AI-Generated Image Detection

2025
Proceedings of the 14th IJCNLPBenchmarkingDeep Learning

N Imanpour, A Borah, S Bajpai, S Dixit, et al.

A benchmark framework for AI-generated image detection that surfaces failure modes and quantifies how detectors degrade as generative models improve.

Contributions

  • Co-authored benchmark framing and evaluation goals aligned with real deployment constraints.
  • Clarified metrics, comparisons, and what constitutes a meaningful detection signal.
  • Helped articulate key breakdown points in current detection pipelines.

Abstract

Introduces the Visual AI Index (V_AI) and a benchmark for evaluating AI-generated image detection, highlighting gaps in current detection pipelines against increasingly capable generative models.

Programmable logic controller (PLC) hardware

Cross-Compatible Encryption Adapter for Securing Legacy Modbus Devices

2025
COMSNETSSCADASystemsSecurity•
DOI

Shreyas Dixit; T. S. Sreeram

A retrofit-friendly hardware adapter that brings modern encryption to legacy Modbus loops without requiring costly system overhauls.

Contributions

  • Worked on retrofit-first design: preserve compatibility while improving confidentiality.
  • Helped shape the threat model and practical constraints (cost, deployment, interoperability).
  • Contributed to system-level evaluation and writing emphasizing deployability.

Abstract

SCADA systems often rely on legacy protocols like Modbus that lack built-in security. This work proposes a cost-efficient, cross-platform hardware adapter to bring modern encryption to legacy Modbus devices with minimal operational disruption.